MindOmni：利用 RGPO 释放视觉语言模型中的推理生成能力
MindOmni: Unleashing Reasoning Generation in Vision Language Models with RGPO
AI总结：
本文提出统一多模态大语言模型 MindOmni，通过 decoder-only 扩散模块、CoT 监督微调及 RGPO 强化学习实现细粒度推理生成，并在理解与生成基准上超越现有模型。
摘要：
近期的文本到图像系统在处理多模态输入和复杂推理任务方面存在局限。我们提出 MindOmni，这是一个统一的多模态大语言模型，通过强化学习引入推理生成来应对这些挑战。MindOmni 采用三阶段训练策略：i）设计一个带有 decoder-only 扩散模块的统一视觉语言模型；ii）使用 Chain-of-Thought（CoT）指令数据进行监督微调；iii）采用我们提出的 Reasoning Generation Policy Optimization（RGPO）算法，利用多模态反馈有效指导策略更新。实验结果表明，MindOmni 优于现有模型，在理解和生成基准测试上均取得出色表现，同时展现出先进的细粒度推理生成能力，尤其是在数学推理指令方面。所有代码将在 https://github.com/TencentARC/MindOmni 公开发布。
英文摘要：
Recent text-to-image systems face limitations in handling multimodal inputs and complex reasoning tasks. We introduce MindOmni, a unified multimodal large language model that addresses these challenges by incorporating reasoning generation through reinforcement learning. MindOmni leverages a three-phase training strategy: i) design of a unified vision language model with a decoder-only diffusion module, ii) supervised fine-tuning with Chain-of-Thought (CoT) instruction data, and iii) our proposed Reasoning Generation Policy Optimization (RGPO) algorithm, utilizing multimodal feedback to effectively guide policy updates. Experimental results demonstrate that MindOmni outperforms existing models, achieving impressive performance on both understanding and generation benchmarks, meanwhile showcasing advanced fine-grained reasoning generation capabilities, especially with mathematical reasoning instruction. All codes will be made public at https://github.com/TencentARC/MindOmni
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
